View Seminar Video
Abstract
The Pen Computing Group at the Thomas J. Watson Research Center has developed ThinkScribe, a writing tablet whose “smart paper” medium enables writers to save all their handwritten notes and drawings in their computers. Large vocabulary (40,000 words) unconstrained handwriting recognition technology, also developed at Watson can transform the digitized text into standard ASCII characters. The device has potential use in a number of different fields.
In this talk, I will present a brief overview of ThinkScribe and discuss some of the challenges faced by Pen computing in the past and how ThinkScribe addresses some of them. I will then move on to talk about the handwriting recognition engine that forms a key component of this device. The recognition engine uses hidden Markov models to model individual characters. A combination of the character models and smart search strategies result in a real time large vocabulary handwriting recognition system. Although the system offers a user the convenience of operation `out of the box’, i.e. without any training on the part of the user, there is a significant improvement in accuracy when trained for an individual’s handwriting. In this talk I will briefly talk about strategies that are used to adapt the writer-independent models to a user’s writing style.
The talk will conclude with a list of open research areas in large vocabulary unconstrained online handwriting recognition and future directions of on-line pen computing.